app.py

import os
import pandas as pd
import torch
import datetime
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
from torch import optim
from torch.optim import AdamW
from transformers import TrOCRProcessor, VisionEncoderDecoderModel, Seq2SeqTrainer, Seq2SeqTrainingArguments, default_data_collator
from sklearn.model_selection import train_test_split
from dataclasses import dataclass
from PIL import Image
from torchvision import transforms
import matplotlib.pyplot as plt
from tqdm.notebook import tqdm
block_plot = False
plt.rcParams['figure.figsize'] = (12, 9)
os.environ["TOKENIZERS_PARALLELISM"] = "true"

model_type="large" #small|base|large

@dataclass(frozen=True)
class TrainingConfig:
  BATCH_SIZE:    int = 15
  EPOCHS:        int = 20
  LEARNING_RATE: float = 0.00002

@dataclass(frozen=True)
class ModelConfig:
  MODEL_NAME: str = 'microsoft/trocr-'+model_type+'-printed'


# Charger le modèle entraîné à partir du fichier .pt
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
trained_model = VisionEncoderDecoderModel.from_pretrained(ModelConfig.MODEL_NAME)
trained_model.load_state_dict(torch.load('ocr_model_large_2024-08-28_14_44.pt', map_location=torch.device('cpu')))
trained_model.to(device)
trained_model.eval()

import gradio as gr
import torch
from PIL import Image
from transformers import VisionEncoderDecoderModel, TrOCRProcessor
import numpy as np

processor = TrOCRProcessor.from_pretrained(ModelConfig.MODEL_NAME)
# Fonction d'inférence
def ocr(image):
    image = Image.fromarray(np.array(image))  # Assurez-vous que l'image est au format PIL
    pixel_values = processor(image, return_tensors='pt').pixel_values.to(device)
    generated_ids = trained_model.generate(pixel_values)
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    return generated_text

# Créer l'interface Gradio
iface = gr.Interface(
    fn=ocr,
    inputs=gr.Image(type="pil", label="Upload Image",height=300),
    outputs=gr.Textbox(label="Extracted Text"),
    title="OCR Text Extraction",
    description="Upload an image to extract text using TrOCR model."
)
iface.launch()